Colabfold install on aarch64 with GPU support

[wahabk]

Expected Behavior

Colabfold should be installable on aarch64. I am working on porting this to a GraceHopper (GH200) system but there are x86 dependencies. The main issue is tensorflow-cpu. Tensorflow only builds wheels for aarch64 without GPU support https://www.tensorflow.org/install/pip

Furthermore there are x86 conda dependencies in the Dockerfile and in localcolabfold:

bash Mambaforge-Linux-x86_64.sh -bfp /usr/local && \

Mainly creating this issue to understand whether aarch64 support is planned and how I can help with this. Conda dependencies can be fixed with $(uname -m).

Current Behavior

ERROR: Could not find a version that satisfies the requirement tensorflow-cpu<3.0.0,>=2.12.1; sys_platform != "darwin" (from colabfold) (from versions: none)
ERROR: No matching distribution found for tensorflow-cpu<3.0.0,>=2.12.1; sys_platform != "darwin"

Steps to Reproduce (for bugs)

Please make sure to reproduce the issue after a "Factory Reset" in Colab.
If running locally ypdate you local installation colabfold_batch to the newest version.
Please provide your input if you can share it.

pip install -q --no-warn-conflicts "colabfold[alphafold-minus-jax] @ git+https://github.com/sokrypton/ColabFold"

ColabFold Output (for bugs)

Please make sure to also post the complete ColabFold output. You can use gist.github.com for large output.

NA can't install colabfold.

Context

Providing context helps us come up with a solution and improve our documentation for the future.

Your Environment

Include as many relevant details about the environment you experienced the bug in.

• Git commit used
• If you run it on a local system. Please add the server specifications
• Operating system and version:

Colabfold Version 1.5.5

$ uname -a
Linux nid001041 5.14.21-150500.55.31_13.0.53-cray_shasta_c_64k #1 SMP Mon Dec 4 22:56:47 UTC 2023 (03d3f83) aarch64 aarch64 aarch64 GNU/Linux
$ cat /etc/os-release 
NAME="SLES"
VERSION="15-SP5"
VERSION_ID="15.5"
PRETTY_NAME="SUSE Linux Enterprise Server 15 SP5"
ID="sles"
ID_LIKE="suse"
ANSI_COLOR="0;32"
CPE_NAME="cpe:/o:suse:sles:15:sp5"
DOCUMENTATION_URL="https://documentation.suse.com/"

[milot-mirdita]

I've been working on getting this running but I am stuck on the conda tensorflow package not offering ARM builds (conda-forge/tensorflow-feedstock#136). You can still install ColabFold manually with pip. Pip has aarch64 builds for jaxlib.

[wahabk]

Thanks @milot-mirdita. I'm aware of that feedstock issue and will try to conda-build.

Could you please clarify what you mean by "install manually with pip"? I'm still getting the error:

$ pip install -q --no-warn-conflicts 'colabfold[alphafold] @ git+https://github.com/sokrypton/ColabFold'
ERROR: Could not find a version that satisfies the requirement tensorflow-cpu<3.0.0,>=2.12.1; sys_platform != "darwin" (from colabfold) (from versions: none)
ERROR: No matching distribution found for tensorflow-cpu<3.0.0,>=2.12.1; sys_platform != "darwin"

[wahabk]

Thanks for the advice and merging the PR!